[Edaily Reporter Kim Hyun-ah ] LG Uplus(032640)has partnered with OptAI, a company specializing in AI model optimization, to develop technology that enhances the operational efficiency of AI services.

As AI services proliferate, GPU and power consumption are rising rapidly. Operating AI with fewer resources while maintaining the same performance has become a critical challenge.

The technology both companies are focusing on is “token optimization.”

A token is a unit of data processed by AI as it understands a question and generates a response. Token optimization is a technology that either streamlines AI models or improves computational methods to enable the same GPU to process more requests.

LG Uplus will validate and apply the technology while operating actual AI services. OptAI will be responsible for developing technology to run AI models more efficiently and at a faster pace.

Through this collaboration, the two companies plan to reduce GPU and power consumption while improving AI response speeds and service quality.

Previously, the two companies also collaborated on on-device AI. LG Uplus applied lightweighting technology to a small language model based on ExaOne, enabling it to run on the NPU—a dedicated AI chip in smartphones.

As a result, they reduced power consumption by 78% and model size by 82% while maintaining performance comparable to that of traditional CPU-based methods.

This time, they are expanding the scope of their collaboration to server GPU environments. They are conducting joint research on technologies to optimize the computational processes of AI models so that a single GPU can handle more user requests when operating large-scale AI services.

This research has already yielded results. The two companies optimized the process by which AI generates responses to suit real-world service environments, increasing the number of tokens that can be processed on a single GPU by up to four times compared to previous levels.

OptAI is an AI optimization startup founded in November 2021 and led by CEO Lee Jae-ho, a former LGELECTRONICS executive. The company possesses technology to lightweight and optimize AI models for hardware such as NPUs, GPUs, and CPUs, and is collaborating with LGELECTRONICS, LG Uplus, and others. It is expanding its business into smartphones, automobiles, and robots through its on-device AI optimization platform, “OptHancer.”

The two companies also co-hosted an “Efficient AI Technology Seminar” on the 1st. Participants including FuriosaAI, Vessel AI, the Korea Electronics Technology Institute (KETI), and Hanyang University shared insights on AI model optimization, AI semiconductors, and real-world service application cases.

LG Uplus plans to gradually apply the acquired technology to its own AI services and large-scale AI infrastructure.

Lee Sang-yeop, CTO of LG Uplus, said, “AI competitiveness goes beyond simply improving performance; it depends on how efficiently we can process as many tokens as possible with the same resources,” adding, “We will expand our research on token optimization technology in collaboration with various partners.”

Lee Jae-ho, CEO of OptAI, remarked, “In the era of generative AI, ensuring operational efficiency is just as important as achieving high performance,” and added, “We will further advance AI optimization technologies that can be validated in real-world service environments.”